Results and Analyses
28/12/23
Tables of tackle values
Real tackle frame (Def 1-10)
Hypothetical tackle frame (Def 2-11)
Tackle gains
Possible issues
Gain analyses
gain and gain2 significantly positive (intuitively this should be the case)
t.test(tackle_vals$`ep_tackle_def2-11`,tackle_vals$`ep_tackle_def1-10`,paired = TRUE)##
## Paired t-test
##
## data: tackle_vals$`ep_tackle_def2-11` and tackle_vals$`ep_tackle_def1-10`
## t = 68.815, df = 11899, p-value < 2.2e-16
## alternative hypothesis: true mean difference is not equal to 0
## 95 percent confidence interval:
## 0.1753847 0.1856692
## sample estimates:
## mean difference
## 0.1805269t.test(tackle_vals$`ep_tackle_def2-11`,tackle_vals$ep_tackle_real_eop,paired = TRUE)##
## Paired t-test
##
## data: tackle_vals$`ep_tackle_def2-11` and tackle_vals$ep_tackle_real_eop
## t = 48.535, df = 11899, p-value < 2.2e-16
## alternative hypothesis: true mean difference is not equal to 0
## 95 percent confidence interval:
## 0.2582366 0.2799730
## sample estimates:
## mean difference
## 0.2691048
Yardlines analyses
Tackle yardlines vs true EOP yardlines (from pbp data).
I thought they should be quite similar…
yardline_diff <- tackle_info_real2$tackle_yardline-tackle_info_real2$true_yardline
summary(yardline_diff)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -65.7300 0.1700 0.8900 0.7667 1.8400 22.2800
Examples of what may happen:
- gameplayId = 2022092500586: here apparently tackle led to fumble which was recovered for a TD from the defense (i.e. in this case tackle was very good)
## playDescription
## 1 (5:44) A.Kamara right tackle to NO 43 for -2 yards (F.Luvu). FUMBLES (F.Luvu), touched at NO 42, RECOVERED by CAR-M.Haynes at NO 44. M.Haynes for 44 yards, TOUCHDOWN. Penalty on NO-N.Vannett, Offensive Holding, declined.
## playResult
## 1 -45## gameplayId gameId playId ep_tackle_def2-11 ep_tackle_def1-10 info
## 1 2.022093e+12 2022092500 586 1.209399 1.146985 1
## gain ep_tackle_real_eop info2 gain2 tackler_id
## 1 0.06241438 -3.02037 1 4.22977 46527- gameplayId = 2022091101382: here penalty on the play (offensive holding -> -10 yards; i.e. tackle was not really necessary)
## playDescription
## 1 (8:46) (No Huddle) B.Mayfield pass short right to C.McCaffrey to CLV 44 for 16 yards (J.Owusu-Koramoah; A.Walker). PENALTY on CAR-I.Thomas, Offensive Holding, 10 yards, enforced at CAR 40 - No Play.
## playResult
## 1 -10## gameplayId gameId playId ep_tackle_def2-11 ep_tackle_def1-10 info
## 1 2.022091e+12 2022091101 382 3.790748 3.656282 1
## gain ep_tackle_real_eop info2 gain2 tackler_id
## 1 0.1344667 0.01406716 1 3.776681 53481- gameplayId = 20220925061332: here again penalty on the play (defensive unnecessary roughness -> -15 yards; i.e. tackle was still necessary, but gain2 is just wrong)
## playDescription
## 1 (10:13) J.Burrow pass deep right to T.Higgins to NYJ 30 for 45 yards (A.Gardner; L.Joyner). CIN-T.Higgins was injured during the play. PENALTY on NYJ-L.Joyner, Unnecessary Roughness, 15 yards, enforced at NYJ 30.
## prePenaltyPlayResult playResult
## 1 45 60## gameplayId gameId playId ep_tackle_def2-11 ep_tackle_def1-10 info
## 1 2.022093e+13 2022092506 1332 4.585782 3.467349 1
## gain ep_tackle_real_eop info2 gain2 tackler_id
## 1 1.118433 4.71303 0 -0.1272486 54469Eliminate penalty play
This may help
Still a lot of negative yardline differences: First inspections show that most of them are due to fumbles with recoveries of the defense (here EP model is wrong and thus gain 2 not accurate)
More problematic: positive yardline difference! Not yet sure what happens here (example play: uId == “2022091809_369”, see also weird looking play2 below)
Issues with some plays
## gameplayId gameId playId ep_tackle_def2-11 ep_tackle_def1-10 info
## 1 2.022092e+12 2022091807 191 2.38869827 3.501782 0
## 2 2.022092e+12 2022091900 256 5.01469894 6.304934 0
## 3 2.022092e+13 2022091804 2905 -0.07842941 6.092776 0
## 4 2.022101e+13 2022100901 3263 2.74110338 6.135961 0
## gain ep_tackle_real_eop info2 gain2 tackler_id
## 1 -1.113084 3.8399481 0 -1.45124985 43298
## 2 -1.290235 4.8127037 1 0.20199529 54500
## 3 -6.171206 -0.1182011 1 0.03977168 41257
## 4 -3.394858 2.6689868 1 0.07211654 52647Something goes wrong when predicting “real” tackle frame?
## gameplayId gameId playId ep_tackle_def1-10 avg_yg uId
## 1 2.022092e+12 2022091807 191 3.501782 6.951991 2022091807_191
## 2 2.022092e+12 2022091900 256 6.304934 NaN 2022091900_256
## 3 2.022092e+13 2022091804 2905 6.092776 NaN 2022091804_2905
## 4 2.022101e+13 2022100901 3263 6.135961 NaN 2022100901_3263
## tackle_yardline yardline_pred yardline_100 playResult absoluteYardlineNumber
## 1 38.43 31.47801 41 10 51
## 2 22.33 NaN 31 8 41
## 3 85.51 NaN 90 5 20
## 4 41.99 NaN 54 12 64
## true_yardline
## 1 31
## 2 23
## 3 85
## 4 42No Problem when predicting hypothetical tackle frame:
## gameplayId gameId playId ep_tackle_def2-11 avg_yg uId
## 1 2.022092e+12 2022091807 191 2.38869827 6.0626123 2022091807_191
## 2 2.022092e+12 2022091900 256 5.01469894 2.7127785 2022091900_256
## 3 2.022092e+13 2022091804 2905 -0.07842941 0.8169310 2022091804_2905
## 4 2.022101e+13 2022100901 3263 2.74110338 0.8312422 2022100901_3263
## tackle_yardline yardline_pred yardline_100 playResult true_yardline
## 1 38.43 32.36739 41 10 31
## 2 22.33 19.61722 31 8 23
## 3 85.51 84.69307 90 5 85
## 4 41.99 41.15876 54 12 42Example plays
Plot description:
- Similar to plot by Ole.
- purple dotted line: Line of scrimmage
- darkgreen dotted line: real EOP yardline
- solid black (chartreuse) line: estimated average EOP yardline
- upper panel: Real tackle frame with def 1-10
- lower panel: Hypothetical tackle frame with def 2-11
Good looking play
Weird looking play
Tackle at yardline around 38 but EOP yardline and (avg) predicted EOP at around 31?
Weird looking play2
What is happening here? Tackle at yardline 80, EOP yardline at 64, prediction with real frame 63.89, prediction with hypothetical frame 70.89???
Play from above with high (positive) yardline difference
